Sleep Stage Classification for Inter-institutional Transfer Learning

Dongyoung Kim, Yunhee Woo, Jaemin Jeong, Dong-Kyu Kim, Jeong-Gun Lee

Abstract: This paper proposes multi-institutional automatic two-channel sleep stage classification by employing a transfer learning technique for the effective use of sleep stage classification backbone models. Only two channels, EEG (Electroencephalo-gram) and EOG (Electrooculogram), are used for patient's sleep stage classification for the real-time clinical decision support. We use two different datasets obtained from different hospitals in order to show the effectiveness of the proposed inter-institutional transfer learning. The experimental results show that 12.5% higher accuracy is obtained through the inter-institutional transfer learning when compared to the original backbone model.